JP2938653B2 - Quantizer in video coding device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a quantizer for a high-efficiency image coding apparatus which divides an original signal into a plurality of components by a filter bank or an orthogonal transform and encodes the divided signals.

[0002]

2. Description of the Related Art For example, JW Woods and SD O'neil, “Sub
band Coding of Images ”, IEEE Trans. on Acoustics,
Speech, and Processing, vol.ASSP-34, No.5, pp.127
The method described in 8-1288 (1988) uses the variance σ
i ² , distortion Di, code amount Ri, total distortion ΣiDi, total code amount R =
Assuming that ΣRi / Ni and the relationship between the code amount of each component and the distortion is expressed by equation (1), the simultaneous dDi / dRi
= Λ / Ni to determine the code amount to be assigned. Where λ is Lagrange's unconstant number and N
i is a ratio of the total number of data to the number of data of the i component. For example, for a division in which the original data is band-limited to 1/2 frequency domain and then subsampled to 1/2 data number, Is Ni = 2. As a result, equation (2) is obtained.

[0003]

(Equation 1)

[0004]

(Equation 2)

The amount of code to be assigned to each component is determined from the above equation (2). Using equation (1), the distortion can be calculated from the assigned code amount, so that the code amount and distortion of each component can be obtained, and the theoretical limit value of the total distortion when the total code amount is constant is obtained. be able to.

Conventionally, a quantizer has been designed so as to satisfy a predetermined code amount, for example, by assigning a different nonlinear quantizer to each component.

[0007]

However, in the conventional method, the designed quantizer has the following problems.

(1). Since a different nonlinear quantizer is used for each component, the scale of hardware increases.

(2). The code amount-distortion characteristic of the quantizer is
Strictly speaking, the characteristic may deviate from the theoretical limit without following the equation (1).

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and an object of the present invention is to use a quantizer whose hardware is simple to quantify characteristics close to the theoretical limit. It is an object of the present invention to provide a technology capable of realizing a vessel.

The above and other objects and novel features of the present invention will become apparent from the description of the present specification and the accompanying drawings.

[0012]

In order to achieve the above object, the present invention considers the above points, and makes the ratio of the quantization width to each component constant, and obtains the quantization width by numerical calculation. Design the quantizer. That is, the input digital image signal (for example, the time of the digital image signal sequence)
After suppressing redundancy using motion compensated inter-frame prediction
Digital image signal or digital image of still image
The spatial redundancy of the image signal is calculated using multi-stage filter banks (eg,
For example, wavelet transform) or orthogonal transform (for example,
The discrete cosine transform), divided into a plurality of components, each component which is the fractional <br/> split, each quantized in the quantizer in the video coding apparatus for variable length coding the quantized signal, The quantizer calculates a variance value of the signal of each component.
A first means for issuing, and each component calculated by the first means.
The average code amount to the target code amount
A second means for calculating a quantization width having a predetermined value with respect to
And a plurality of linear quantizations provided for each component signal
A vessel, said second means based on the calculated quantization width
Then, the quantization width for each signal of each component is obtained,
Each component using the quantization width of each component
Signals and the most important features that have a linear quantizer for quantizing the.

[0013]

According to the above-mentioned means, since the quantization error is minimized for a predetermined code amount in accordance with the variance of the signal of each divided component, a straight-line quantizer which can be easily implemented in hardware is used. Thus, the code amount-distortion characteristic close to the conventional theoretical limit can be realized. When using general band division,
The above-described characteristics can be realized by performing a simple calculation on a signal to be quantized using a single quantizer.

[0014]

Embodiments of the present invention will be described below in detail with reference to the drawings.

For a scalar-type optimal quantizer that gives a minimum square distortion under a constant entropy condition, T. Berge
r, “Optimum Quantizers and Permutation Codes”, I
EEETrans. Inform. Theory, vol.IT-18, No.6, pp.759-
765 (Nov. 1972) reports that it is almost linear regardless of the distribution of the input signal.

Also, RC Wood: “On optimum Quatizat
ion ”, IEEE Trans. Inform. Theory, vol.IT-5, pp.2
According to the approximation in 48-252 (Mar. 1969), the code amount R of the linear quantizer and the quantization width h have a relationship represented by Expression (3).

[0017]

(Equation 3)

The quantization width for giving the code amount allocated by the conventional method is given by equation (4) from equations (3) and (2).

[0019]

(Equation 4)

That is, an approximate solution is obtained in which the quantization width of each component is proportional to 1 / √Ni (√ means the square root of Ni) irrespective of its variance. In the present invention, in consideration of the above points, the ratio of the quantization width to each component is fixed, and the quantization width is obtained by numerical calculation.

FIG. 1 is a block diagram showing a configuration of a quantization width control circuit according to an embodiment of the present invention, wherein 1 is a dispersion calculation circuit, 2 is a quantization width calculation circuit, and 3 is a linear quantizer. It is.

As shown in FIG. 2, the linear quantizer 3 includes a first ROM (Read Only Memory) 31 and a second ROM (Read Only Memory) 31.
It is composed of OM32. The first ROM 31 stores a value of the quantization width h obtained by the quantization width calculation circuit 2 corresponding to each component of √Ni (Ni is the ratio of the total number of data to the i component: where √ is the square root of Ni. ) (H / √Ni) to determine the actual quantization width hi,
The results are tabulated and recorded in the semiconductor memory device. When the quantization width hi and the component i (the number of data of the component i = the total number of data / Ni), which are the outputs of the first ROM 31, are input, the second ROM 32 has a table for outputting the quantization number of the component i. It is recorded on a semiconductor memory device. That is, the component i in the second ROM 32
The corresponding quantization number is recorded in an address consisting of a numerical value obtained by arranging the data value and the quantization width hi. If such a ROM read circuit is configured, a quantization number can be obtained simply by specifying an address from which data is read. A well-known distributed calculation circuit is used. The quantizer is RO
M is configured as a basic element.

As shown in FIG. 1, the quantization width control circuit according to the present embodiment uses the variance calculation circuit 1 to calculate the variance σ ₁ ² ,
σ ₂ ² σ ₃ ² ,... σ M ² (where M is a subscript of σ) is input to the quantization calculation circuit 2. The quantization calculation circuit 2 calculates the quantization width h in accordance with the flowchart shown in FIG. 3 described later, and inputs the quantization width h to the linear quantizer 3 for each component together with the data of each component.

FIG. 3 is a flowchart showing a quantization width calculation processing procedure in the quantization width calculation circuit 2 shown in FIG.

In FIG. 3, h is the quantization width, Robj is the target code amount, σi is the variance value of the i component, Ri is the code amount of the i component, Ni is the ratio of the total number of data to the i component, hmax
Is a temporary maximum value of the quantization width, hmin is a temporary minimum value of the quantization width, and ε is a threshold value.

The function r (h / √Ni, σ ² ) is a variable h / √Ni, σ
Is expressed by the following equation (5).

[0027]

(Equation 5)

In the quantization width calculation process in the quantization width calculation circuit 2, first, the quantization width h is appropriately determined. For example, h
= 1 and the number of components is M.

As shown in FIG. 3, at step 101, the variance σi of each component i (i = 1, 2, 3,... M) and the data number ratio Ni are known.

The code amount R of the component i represented by the above equation (5)
i is calculated for i = 1, 2, 3,... M. The average code amount R integrated with each component is calculated by the following equation (6).

[0031]

(Equation 6)

In step 102, the average code amount R is compared with the target code amount Robj, and the average code amount R is compared with the target code amount Robj.
If so (NO), in step 103, the quantization width h
Is doubled (h ′ = 2h), and equations (5) and (6) are calculated again. In step 105, the average code amount R and the target code amount R are calculated.
Compare with obj. This operation is repeated until the average code amount R exceeds the target code amount Robj.

If the average code amount R is equal to or larger than the target code amount Robj in step 2 (YES), step 104 is executed.
Then, the quantization width h at this time is stored in the memory as a temporary maximum value hmax. Next, h (= hmax) is multiplied by 1/2.

In step 105, the above equations (5) and (6) are calculated with the quantization width h '' (= h / 2) obtained by multiplying h (= hmax) by 1/2 as the quantization width h. , In step 106,
The average code amount R is compared with the target code amount Robj. If the average code amount R is equal to or smaller than the target code amount Robj (NO), in step 107, h is further multiplied by ２ to obtain the target code amount R. This is repeated until the code amount becomes Robj or more. Usually 1 / once
By simply multiplying by 2, R> Robj. If the average code amount R is equal to or larger than the target code amount Robj (YES), the quantization width h at this time is stored in the memory as a temporary minimum value hmin (step 108).

The quantization width to be obtained is between the maximum value hmax and the minimum value hmin. In step 109, the maximum value hmax
And the average value of the minimum value hmin and the average value is defined as the quantization width h. At this time, the value of R with respect to h is the target value code amount Robj.
When the absolute value error is less than a certain threshold value ε, the obtained h is regarded as the final quantum width.

Otherwise, the quantization width h is changed by the following processing.

In step 110, h = (hmax + hmin)
The average code amount R is obtained by the formulas (5) and (6) assuming that the value is / 2. In step 111, | R-Robj | is compared with the threshold ε, and | R-Robj | is smaller than the threshold ε (|
If R−Robj | <ε: YES), the process ends. |
R-Robj | is larger than the threshold value ε (| R-Robj |>
ε: NO), the average code amount R
Is compared with the target value code amount Robj, and when the average code amount R is smaller than the target value code amount Robj (R <Robj: YES), h is replaced with hmax. Conversely, when the average code amount R is larger than the target value code amount Robj (R> Robj: NO), h is set to h.
Replace with min.

In this processing, when the value of the average code amount R is viewed as an absolute value error with respect to the target value code amount Robj, a certain threshold ε
Repeat until:

As can be seen from the above description, according to the present embodiment, all the quantizers to be prepared are linear quantizers, and the hardware configuration is simpler than in the conventional method.
In the orthogonal transform, since Ni is equal in each component, the same quantizer can be used for each component. In the case of division by the filter banks, the pixel number ratio of each component, since generally it is 2 ² _n, prepared single quantizer, by multiplying 2 _n signals to be quantized The above-described characteristics can be realized.

The 2 ⁿ times of the signal can be easily realized by bit shifting when the signal is represented by an integer or a fixed-point real number, and by adding an exponent part when the signal is represented by a floating-point real number.

Further, DP Garredo and WA Finamore,
“Optimum Bit Allocation Procedure for Hierarchica
l Subband Coding of Image ", Proc. IMAGE 'COM 90, p
p.126-129 (Nov. 1990). Using the quantizer designed by the above method for the 7-component signal with different variances in Table 1 quoted from Table 1, In comparison, the difference is small as shown in FIG. 4, and a characteristic close to the logical limit can be realized by the above method. In FIG. 4, D is the variance of the quantization error.

[0042]

[Table 1]

FIG. 5 is a block diagram showing a schematic configuration of an embodiment of a high-efficiency image coding apparatus to which the quantizer of the present invention is applied. Reference numeral 401 denotes a band splitter;
03,..., 50M are quantizers, 601, 602, 60
, 60M are variable length encoders.

The input image is divided into N components by a band divider 401, and quantizers 501, 50 designed by the above-described method.
, 50M, and are variable-length coded by variable-length encoders 601, 602, 603,..., 60M, respectively, and output as compressed code strings.

This embodiment is an encoding apparatus for encoding a still image or intra-frame encoding of a moving image.
When applied to inter-frame coding of a moving image, it can be easily analogized that inter-frame motion compensation prediction is performed, and the prediction error signal may be coded by the configuration shown in the present embodiment. .

Although the present invention has been described in detail with reference to the embodiments, the present invention is not limited to the above embodiments.
It goes without saying that various changes can be made without departing from the gist of the invention.

[0047]

As described above, according to the present invention, a code amount-distortion characteristic close to the conventional theoretical limit can be realized by using a linear quantizer which can be easily implemented in hardware. In addition, when general band division is used, the above-described characteristics can be realized by using a single quantizer and performing a simple calculation on a signal to be quantized.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a quantization width control circuit according to an embodiment to which the present invention is applied;

FIG. 2 is a block diagram showing a configuration of the linear quantizer shown in FIG. 1;

FIG. 3 is a flowchart illustrating a quantization width calculation processing procedure in the quantization width calculation circuit illustrated in FIG. 1;

FIG. 4 is a diagram showing a code amount-distortion characteristic close to a conventional theoretical limit and a code amount-distortion characteristic according to the present embodiment;

FIG. 5 is a block diagram showing a schematic configuration of an embodiment of a high-efficiency image coding apparatus to which the quantizer of the present invention is applied.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Distributed calculation circuit, 2 ... Quantization calculation circuit, 3 ... Linear quantizer, 31 ... 1st ROM, 32 ... 2nd ROM, 401 ...
, 50M... Quantizers for each component, 601, 602, 603,.
60M: Variable length encoder for each component.

Continuation of the front page (56) References JP-A-3-191618 (JP, A) 1990 Annual Meeting of the Institute of Television Engineers of Japan. 119-120 (Study of High Efficiency Coding for Digital VTR (Part 2)-Subband DCT-Switching of Linear Quantization Characteristics According to Block Activity) 1990 IEICE Spring Meeting, 1990 D-268 “Study on quantization in subband region of image” (58) Fields investigated (Int. Cl. ⁶ , DB name) H04N 1/41-1/419 H04N 7/24-7/68

Claims (3)

(57) [Claims] 1. An image processing apparatus comprising:
Redundancy, divided into <br/> plurality components by multi-stage filter bank or orthogonal transform, the divided plurality of components and each quantized, quantization in the video encoding apparatus for variable length coding the quantized signal In the quantizer, the quantizer calculates a variance value of the signal of each component.
The first means and a variance value of the signal of each component calculated by the first means.
Therefore, the average code amount becomes a predetermined value with respect to the target code amount.
A second means for calculating a quantization width, and a plurality of linear quantizers provided for each signal of each component.
And based on the quantization width calculated by the second means.
The quantization width for each signal of each of the components is obtained, and the obtained
Using the quantization width of each signal of each component, the signal of each component is used.
And a linear quantizer for quantizing the signal. 2. The input digital image signal comprises:
A motion compensation frame is used to determine the temporal redundancy of a digital image signal sequence.
Digital image signal after suppression using inter-frame prediction.
The quantizer according to claim 1, wherein 3. The input digital image signal comprises:
A digital image signal of a still image;
The quantizer according to claim 1 .